The results of this study indicate that the FFQ is a reasonably good method for dietary assessment among pregnant women in Spain. The FFQ showed a good degree of reproducibility for most of the foods and nutrient intakes, and an acceptable relative validity for some nutrients of special interest during pregnancy. This validation study was undertaken because a new FFQ has been developed for assessment of the dietary intakes of pregnant women in the INMA-Project, a prospective cohort study of pregnant women and their offspring [9, 10]. On the other hand, this is the first study ever done to validate a FFQ among pregnant women in Spain with a sample size large enough to detect as statistically significant even small correlation coefficients which may be considered as minimally valid (r>0.20).
The correlation coefficients for most of the nutrients and food groups were comparable with those observed in other validation studies of FFQ among pregnant women and other populations [3, 5]. Reproducibility was assessed by comparing the results from the FFQ at two different periods in pregnancy while validity was explored by comparing the nutrient intakes from FFQ and biomarker concentrations of several nutrients in plasma that are potentially important for perinatal research and that are sensitive to dietary intake (carotenoids, folate, vitamin B12 and C, E and retinol). Despite the use of biomarkers as a reference methods may have some limitations since they do not provide a quantitative measure of dietary intake but rather a qualitative indicator and may be more related to recent intake while FFQ to long-term exposure, plasma concentration for some nutrients are sensitive and acceptable indicators of intakes. Therefore, their use for FFQ validation is considered appropriate since the two methods of assessing diet have different sources of error that are unlikely to be correlated with each other [7, 8].
Overall, our FFQ has shown good reproducibility and therefore, it may be considered a reliable dietary assessment method among pregnant women of the INMA Project in Spain. Few studies have been published presenting data on the reproducibility of their FFQs in pregnant women [23–27]. In our study, the average of correlation coefficients between the first and second administration of the FFQ was 0.51 for the intakes of the 29 nutrients considered. Similar results were observed in a study with low-income pregnant women, except for a stronger vitamin A correlation . In a Finnish study, the average of correlation coefficients for all nutrients considered was 0.66, higher than the 0.51 in our study . In a Portuguese study, the estimated average of coefficients for 15 nutrients was also high (0.62), although the interval of the FFQ administration in this study was short, two weeks in the third trimester of pregnancy . In a recently published study with pregnant women in Malaysia , the average of correlation coefficients for nutrients and foods was very high 0.87, which may be related in part to the short time period of administration between FFQ1 and FFQ2 (10 weeks) and to a more stable diet in the third trimester of pregnancy in that country. We applied a 20-week interval which made answer memorization more unlikely, thus avoiding an artificially increased reproducibility. However, a long interval could also be a concern since diet in pregnant women may not be as stable as among non-pregnant women and reproducibility may be compromised by real dietary changes. Although some studies have reported higher nutrient intakes with the second FFQ administration [26, 27], other studies have shown no evidence of major changes in diet during pregnancy . We observed slightly higher intakes for many foods and nutrients during the first period (but not for lycopene, fruits and dairy products), which may support the idea that some changes in diet did occur. Therefore, we explored whether reproducibility correlations were modified by categories of variables such as self-reported changes in diet (no/yes), age, body mass index, country of origin and self-reported vomiting during pregnancy (no/yes); however, correlations remained practically unchanged (data not shown). Despite potential changes in diet during pregnancy, the FFQ showed a satisfactory level of reproducibility for most foods and nutrients, particularly for those more frequently eaten. As far as we know, only two studies have previously explored reproducibility of food groups by FFQ among pregnant women [23, 26]. Although foods were not grouped in the same way in our study, the average of the correlation coefficients were very similar to those observed in the Finnish Study while slightly lower than those reported in other non-pregnant populations .
Regarding the relative validity of the FFQ as assessed by biomarkers, poor correlations were found between dietary intake and plasma concentrations of retinol and α-tocopherol (r=0.05). A low validity has been also found in other studies with pregnant women [28–31] as well as non-pregnant population [4, 32]. The retinol concentrations in plasma is highly regulated by liver stores over a wide range of dietary intakes that can be found mainly in subjects with either severely depleted or highly saturated liver stores . Concerning the poor correlation for vitamin E, it has been suggested that plasma concentration may not be a good marker for usual nutrient intake among pregnant women and that other tissues (e.g., adipose tissue) may better represent usual vitamin E intake . Nevertheless the lack of validity for retinol and vitamin E deserves attention when using this FFQ in the study of diet disease relationships.
The correlation coefficient for lycopene was lower than the observed for other carotene. It has been suggested that in order to increase validity for the assessment of long-term intake, one approach would be to use the average of repeated administrations of a dietary questionnaire . Correlations for lycopene were substantially improved when plasma concentration was compared with the average of the two administrations of the FFQ. Correlation coefficients for α-carotene, β-carotene and β-cryptoxantin were slightly improved when we used the average of the two FFQ, but not so for the others nutrients (data not shown). Poor correlations could be also related to when blood samples and FFQ were collected since the availability of foods with high content of carotenoids may differ substantially according to the season of the year. When the analysis was performed according to the season of the year, lycopene correlations for samples taken in spring and summer were significantly higher than the other two seasons or the overall. For retinol, we observed a better correlation in winter. Thus, validity of the FFQ may be even higher than shown in Table 4 as correlations were improved when time frame and season between nutrient intakes and plasma concentrations were matched more properly in the analysis.
Despite the fact that some correlations were low, our results support that the FFQ may be valid to estimate of long-term intakes and acceptably classify women according to their intakes for relevant nutrients. In fact, correlations in this study were slightly higher than those found in a previous study we performed using a similar FFQ in an elderly population  and in agreement with other studies in pregnant as well as non-pregnant populations [3–5].
On the other hand, the correlation coefficients were substantially improved for folate (from r=0.12 to r=0.53) and to lesser extent for vitamin B12 (from r=0.08 to r=0.12) and vitamin C (from r=0.18 to r=0.20) when dietary supplement use was considered. In Spain, the contribution of vitamin B12 from supplements to the total intake of this vitamin is probably low although it may be included in some folic acid and multivitamin supplements, and therefore may influence the total dietary intake of this vitamin in some periods of pregnancy [34, 35].
Other methods of dietary assessment such as food diaries or 24 h recalls are commonly used as a reference in validation studies of FFQ , and this may be a limitation of this study. However, the use of these methods was problematic and unfeasible in our study because of the large sample size and the considerable proportion of women with low educational level and with a country of origin different from Spain.
Therefore, we only explored relative validity by comparing nutrient intakes with their respective concentrations in plasma. Nutrient levels in plasma may be influenced not only by dietary intake but also by external factors such as the food matrix, the food preparation and by host factors such as gender, smoking status and BMI; however, when we stratified the analysis for some of these variables, correlations did not change and showed acceptable degree of validity still.
In conclusion, our findings show that reproducibility and validity of the FFQ assessed in this study using biological markers were acceptable and comparable with the results of earlier studies. We conclude that our FFQ is a good method for assessing intake of several relevant nutrients during pregnancy.